Skip to main content
IBM Quantum Platform

pauli_feature_map

class qiskit.circuit.library.pauli_feature_map(feature_dimension, reps=2, entanglement='full', alpha=2.0, paulis=None, data_map_func=None, parameter_prefix='x', insert_barriers=False, name='PauliFeatureMap')

GitHub

Bases:

The Pauli expansion circuit.

The Pauli expansion circuit is a data encoding circuit that transforms input data x⃗∈Rn\vec{x} \in \mathbb{R}^n, where nn is the feature_dimension, as

UΦ(x⃗)=exp⁡(i∑S∈IϕS(x⃗)∏i∈SPi).U_{\Phi(\vec{x})}=\exp\left(i\sum_{S \in \mathcal{I}} \phi_S(\vec{x})\prod_{i\in S} P_i\right).

Here, SS is a set of qubit indices that describes the connections in the feature map, I\mathcal{I} is a set containing all these index sets, and Pi∈{I,X,Y,Z}P_i \in \{I, X, Y, Z\}. Per default the data-mapping ϕS\phi_S is

ϕS(x⃗)={xi if S={i}∏j∈S(π−xj) if ∣S∣>1.\phi_S(\vec{x}) = \begin{cases} x_i \text{ if } S = \{i\} \\ \prod_{j \in S} (\pi - x_j) \text{ if } |S| > 1 \end{cases}.

The possible connections can be set using the entanglement and paulis arguments. For example, for single-qubit ZZ rotations and two-qubit YYYY interactions between all qubit pairs, we can set:

circuit = pauli_feature_map(..., paulis=["Z", "YY"], entanglement="full")

which will produce blocks of the form

┌───┐┌─────────────┐┌──────────┐                                            ┌───────────┐
┤ H ├┤ P(2.0*x[0]) ├┤ RX(pi/2) ├──■──────────────────────────────────────■──┤ RX(-pi/2) ├
├───┤├─────────────┤├──────────┤┌─┴─┐┌────────────────────────────────┐┌─┴─┐├───────────┤
┤ H ├┤ P(2.0*x[1]) ├┤ RX(pi/2) ├┤ X ├┤ P(2.0*(pi - x[0])*(pi - x[1])) ├┤ X ├┤ RX(-pi/2) ├
└───┘└─────────────┘└──────────┘└───┘└────────────────────────────────┘└───┘└───────────┘

The circuit contains reps repetitions of this transformation.

Please refer to z_feature_map() for the case of single-qubit Pauli-ZZ rotations and to zz_feature_map() for the single- and two-qubit Pauli-ZZ rotations.

Examples

>>> prep = pauli_feature_map(2, reps=1, paulis=["ZZ"])
>>> print(prep)
     ┌───┐
q_0: ┤ H ├──■──────────────────────────────────────■──
     ├───┤┌─┴─┐┌────────────────────────────────┐┌─┴─┐
q_1: ┤ H ├┤ X ├┤ P(2.0*(pi - x[0])*(pi - x[1])) ├┤ X ├
     └───┘└───┘└────────────────────────────────┘└───┘
>>> prep = pauli_feature_map(2, reps=1, paulis=["Z", "XX"])
>>> print(prep)
     ┌───┐┌─────────────┐┌───┐                                            ┌───┐
q_0: ┤ H ├┤ P(2.0*x[0]) ├┤ H ├──■──────────────────────────────────────■──┤ H ├
     ├───┤├─────────────┤├───┤┌─┴─┐┌────────────────────────────────┐┌─┴─┐├───┤
q_1: ┤ H ├┤ P(2.0*x[1]) ├┤ H ├┤ X ├┤ P(2.0*(pi - x[0])*(pi - x[1])) ├┤ X ├┤ H ├
     └───┘└─────────────┘└───┘└───┘└────────────────────────────────┘└───┘└───┘
>>> prep = pauli_feature_map(2, reps=1, paulis=["ZY"])
>>> print(prep)
     ┌───┐┌──────────┐                                            ┌───────────┐
q_0: ┤ H ├┤ RX(pi/2) ├──■──────────────────────────────────────■──┤ RX(-pi/2) ├
     ├───┤└──────────┘┌─┴─┐┌────────────────────────────────┐┌─┴─┐└───────────┘
q_1: ┤ H ├────────────┤ X ├┤ P(2.0*(pi - x[0])*(pi - x[1])) ├┤ X ├─────────────
     └───┘            └───┘└────────────────────────────────┘└───┘
>>> from qiskit.circuit.library import EfficientSU2
>>> prep = pauli_feature_map(3, reps=3, paulis=["Z", "YY", "ZXZ"])
>>> wavefunction = EfficientSU2(3)
>>> classifier = prep.compose(wavefunction)
>>> classifier.num_parameters
27
>>> classifier.count_ops()
OrderedDict([('cx', 39), ('rx', 36), ('u1', 21), ('h', 15), ('ry', 12), ('rz', 12)])

References:

[1] Havlicek et al. Supervised learning with quantum enhanced feature spaces, Nature 567, 209-212 (2019).

Parameters

Return type

QuantumCircuit

Was this page helpful?
Report a bug or request content on GitHub.